Administrative Actionability in Public-Sector AI Governance: A Task-Specific Documentary Assessment

• In Event: AI Governance and Public Capacity in an Uneven Policy Landscape

Abstract

Public agencies increasingly govern artificial intelligence (AI) through procurement rules, oversight requirements, and administrative procedures. Yet the existence of formal rules, or the deployment of an automated system, does not establish how a particular task is governed. This paper develops administrative actionability as a task- and period-specific documentary assessment of three institutional conditions: an authoritative classification of the governed activity, an identifiable authority responsible for maintaining its rules, and an operative procedure linking the task to consequential review or action. The approach distinguishes formally available procedures from documented recurrent operation and withholds judgment when records are insufficient. Four U.S. episodes involving generative-AI procurement, child-welfare screening, predictive policing, and unemployment-benefits administration illustrate the framework. Their contrasting records illustrate why formal requirements should not be equated with routine compliance, or deployment with clear rule-maintaining authority. The assessment also distinguishes AI-specific uses from administrative automation when AI involvement is unverified. It describes documented institutional arrangements, not whether those arrangements are legal, fair, effective, or legitimate. The paper offers a focused way to examine public-sector AI governance while recognizing that the framework’s reliability and additional analytical value require independent testing.

Author

• Sina Jangjoo, The University of Texas at Tyler
  Presenting Author